EE ZOOM Seminar: FSCDiff: Few-Shot Object Counting with Diffusion Probabilistic Models

Electrical Engineering Systems ZOOM Seminar

Speaker: Yael Grinstein

M.Sc. student under the supervision of Dr. Hadar Averbuch-Elor and Prof. Lior Wolf

Wednesday, 4^th September 2024, at 15:00

Join Zoom Meeting
https://us05web.zoom.us/j/84213774368?pwd=NxPJfZtnRNlmpT8Nt4IbbPsD2Tn9HP.1
Meeting ID: 842 1377 4368
Passcode: nJX6uf
 

FSCDiff: Few-Shot Object Counting with Diffusion Probabilistic Models

Abstract

Few-shot object counting, requiring accurate object enumeration from limited training examples, remains a formidable challenge. Traditional methods often struggle to generalize to unseen object categories due to their reliance on complex feature engineering and regression-based pipelines. This work introduces a novel approach that harnesses the generative power of diffusion models to directly estimate object density maps. By conditioning the diffusion process on query images and a small set of exemplar images, our model learns to progressively refine a noisy image into a precise representation of object distribution. This end-to-end approach eliminates the need for intermediate feature extraction and matching stages, leading to improved performance and generalization. Our method significantly outperforms state-of-the-art techniques on the challenging FSC147 benchmark, demonstrating the potential of diffusion models to revolutionize few-shot object counting.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בצ'ט